Intel announces new Lunar Lake chip series with improved AI processor

Intel pulled back the curtain on its latest series of processors, codenamed Lunar Lake, during Intel CEO Pat Gelsinger’s Computex keynote on Wednesday. The new chip will launch later this year, powering thin and light laptops in time for the 2024 holiday season.

The new system-on-a-chip (SoC) introduces a number of innovations compared to current generation Meteor Lake SoCs, such as Intel Core and Core Ultra processors, with Intel claiming better power efficiency, a 4x increase in power processing NPU, a 50% faster Intel Arc GPU based on its next-generation Arc Battlemage architecture, and on-chip memory for a smaller footprint and faster memory access.

Lunar Lake is also the first Intel chip series to meet the minimum requirements of Microsoft’s PC Copilot+ standards, including more than 40 trillion operations per second for its NPU (TOPS) and 16 GB of memory.

The new SoC features a highly redesigned architecture that introduces a series of small changes to the operation of its processors that collectively achieve significant performance improvements without using more power than current Intel Core Ultra chips, and which could achieve the same performance as Intel Meteor Lake. chips with a third of the power, significantly increasing the potential battery life of laptops equipped with the new chips.

I haven’t tested any of these claims myself, so they should all be taken with a grain of salt until reviewers are able to get their hands on new laptops with Lunar Lake chips, but in the part of an extended technical visit before Computex. In this announcement, Intel engineers gave a pretty comprehensive overview of the changes that Lunar Lake will introduce and they are quite profound, even if the individual changes don’t shake things up much on their own.

New core architectures introduced

The introduction of Intel Lunar Lake at Computex 2024

(Image credit: Avenir / John Loeffler)

Lunar Lake’s biggest innovation is the introduction of a new hybrid core architecture with Lion Cove performance cores and Skymont efficiency cores. These cores are a major driver of Intel’s claimed energy efficiency improvement for Lunar Lake, with a number of small changes to the way instructions are processed by the processor, the introduction of a new layer of cache memory and a doubling of cache memory. number of efficiency cores on Meteor Lake.

Another major change is the loss of hyperthreading capability. This is when a processing core is capable of executing two instructions per clock cycle instead of one, although Intel says the optimization made to Lunar’s Lion Cove performance cores Lake more than makes up for the loss of hyperthreading, which comes with its own processing overhead.

All of these changes, taken together, add up to efficiency cores that Intel says deliver the same performance as current-generation Meteor Lake chips at a third of the power, and up to four times the performance at full power. maximum thanks to the addition of two additional efficiency cores. As for the performance cores, Intel claims that Lunar Lake’s Lion Cove performance cores are capable of achieving 50% higher performance for the same power than Meteor Lake’s Redwood Cove cores in single-threaded performance, and around 5 % higher performance with a single-threaded Lion. Cove Core than Meteor Lake’s multithreaded performance cores.

Additionally, Intel says its Efficiency Cores are mature enough to be able to handle more demanding tasks than previous generations of Efficiency Cores, meaning workloads that were previously assigned to higher performance cores can now be assigned to less powerful efficiency cores. increasing the overall efficiency of the chip, which should help extend the battery life of a Lunar Lake-equipped laptop to be competitive with Arm-based laptops like those powered by the new Snapdragon from Qualcomm and the M3 and M4 chips from Apple.

Expanded NPU and next-generation Intel Arc graphics

Other notable advancements come from the two other major components of the Lunar Lake SoC, namely the Intel NPU 4 neural processor and the Intel Xe2 graphics architecture, based on Intel Arc Battlemage.

Starting with the GPU, the next-gen Arc graphics come with 8 Xe execution cores, 8 ray tracing cores, a dedicated matrix AI engine capable of 67 TOPS and promise up to 50% graphics performance. superior compared to Meteor Lake’s Arc Alchemist-based architecture. .

The NPU, meanwhile, quadruples the performance of Intel Meteor Lake’s NPU 3 neural processor from 12 TOPS to 48 TOPS thanks to four additional neural processing cores and other architectural improvements.

For those keeping count, that’s a total platform TOPS of 120 between GPU, NPU, and CPU, and when Lunar Lake laptops roll out in Q3 and Q4 of this year, Windows Copilot+ Runtime and others Developer APIs will allow applications manufacturers to leverage all three processors as needed to better optimize AI tasks.

It remains to be seen whether Intel Lunar Lake delivers on the promise of PC AI this year, but Intel projects great confidence that by the end of the year it will take a sizable lead in the PC AI market, regardless of whatever form this market will ultimately take. .